espnet3.systems.base.system.BaseSystem

source

class espnet3.systems.base.system.BaseSystem(train_config: DictConfig | None = None, infer_config: DictConfig | None = None, measure_config: DictConfig | None = None, stage_log_mapping: dict | None = None)

Bases: object

Base class for all ESPnet3 systems.

Each system should implement the following: : - create_dataset()

• train()
• infer()
• measure()
• publish()

This class intentionally does NOT implement: : - DAG

• dependency checks
• caching

All behavior is config-driven.

• Parameters:
  • train_config (DictConfig | None) – Training configuration.
  • infer_config (DictConfig | None) – Inference configuration.
  • measure_config (DictConfig | None) – Measurement configuration.
  • stage_log_mapping (dict | None) – Optional overrides for stage log path resolution. Keys are stage names; values are dotted attribute paths (e.g., "train_config.exp_dir") or lists/tuples of such paths (first non-empty value wins).

Stage log mapping (base defaults): : Stage | Path reference | <br/> <br/>

|

<br/> —
<br/>

|

<br/> — | | create_dataset | train_config.recipe_dir | | collect_stats | train_config.stats_dir | | train | train_config.exp_dir | | infer | infer_config.inference_dir | | measure | measure_config.inference_dir | | pack_model | train_config.exp_dir | | upload_model | train_config.exp_dir |

Any stage missing from the mapping (or resolving to None) falls back to the default log directory: train_config.exp_dir when available, otherwise <cwd>/logs.

Examples

Override a subset of stage log paths: : ``` ``

<br/>

`

<br/>
python
system = BaseSystem(
<br/>
> train_config=train_cfg,
> infer_config=infer_cfg,
> measure_config=measure_cfg,
> stage_log_mapping={
<br/>
> > “infer”: “train_config.exp_dir”,
> > “measure”: “train_config.exp_dir”,
<br/>
> },

Initialize the system with optional stage configs.

<div class='custom-h4'><p>collect_stats<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Collect statistics needed for training.

<div class='custom-h4'><p>create_dataset<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Create datasets using the configured stage.

<div class='custom-h4'><p>infer<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Run inference on the configured datasets.

<div class='custom-h4'><p>measure<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Compute evaluation metrics from hypothesis/reference outputs.

<div class='custom-h4'><p>pack_demo<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Pack demo artifacts into a runnable demo bundle.

<div class='custom-h4'><p>pack_model<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Pack model artifacts into an espnet3 bundle.

<div class='custom-h4'><p>publish<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Publish artifacts from the experiment.

<div class='custom-h4'><p>train<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Train the system model.

<div class='custom-h4'><p>upload_demo<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Upload demo bundle to HuggingFace Spaces (stub).

<div class='custom-h4'><p>upload_model<span class='small-bracket'>(*args, **kwargs)</span></p></div>


Upload model bundle to HuggingFace.